Wide area measurement system data time mark abnormity identification method and system

Abstract

The invention relates to the technical field of power systems, in particular to a wide area measurement system data time mark anomaly identification method and system, and the method comprises the following steps: S1, data collection and preprocessing; s2, constructing a time sequence relative variation matrix; s3, establishing an error data time sequence feature strategy table; s4, comparing and judging that the time marks are abnormal; and S5, carrying out abnormity positioning and alarming. According to the method, the time sequence relative variation matrix fused with the fractal features is constructed, and a composite similarity calculation method based on combination of DTW and cosine similarity is adopted to be compared with a predefined strategy table, so that whether the data time scale is abnormal or not and the type of the abnormality can be quickly and accurately judged; compared with a traditional method, the method has the advantages that the judgment speed and accuracy are greatly improved, and the false alarm rate of a WAMS-based power grid fault intelligent diagnosis and accident handling auxiliary decision-making system is reduced.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to a method and system for identifying time-scale anomalies in data from a wide-area measurement system. Background Technology

[0002] The Wide Area Measurement System (WAMS), relying on synchronous phasor technology, enables high-precision synchronous phasor measurement, high-speed communication, and rapid response across different locations, providing core technical support for real-time monitoring of dynamic processes in large-span power grids. This system collects key electrical quantities such as voltage, current, power, and frequency through phasor measurement units (PMUs) distributed throughout the network, and uses the Global Positioning System (GPS) to convert the data to a unified time reference coordinate, achieving synchronous time-stamped recording of phasor information at different locations within the power system.

[0003] In actual operation, WAMS data time-stamp anomalies occur frequently. Due to the PMU's high dependence on GPS signals and the stringent quality requirements of the data transmission channel, errors and data loss in the collected data are prone to occur, leading to time-stamp anomalies. Such abnormal data can seriously affect the functions of various applications based on WAMS monitoring data. In power grid fault diagnosis scenarios, it may cause missed or incorrect fault diagnoses, reduce the accuracy of online diagnosis, and interfere with dispatchers' accurate judgment of accidents.

[0004] Current methods for detecting abnormal WAMS data mainly fall into two categories: traditional state estimation methods and modern data mining methods. Traditional methods, when applied to high-sampling-rate WAMS data (typically ≥25 frames / s), suffer from problems such as high computational cost, insufficient real-time performance, and susceptibility to residual contamination, complex solutions to nonlinear equations, and misdiagnosis of multiple erroneous data. While data mining methods have seen improvements, they lack specialized techniques for identifying time-scaled anomalies. Existing processing methods often employ simple approaches such as adjacent data filling, lacking a comprehensive and accurate time-scaled anomaly identification system. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying time-stamp anomalies in data of a wide-area measurement system, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for identifying time-stamp anomalies in data from a wide-area measurement system includes the following steps:

[0008] Step S1: Data Acquisition and Preprocessing;

[0009] Step S2: Construct a time-series relative change matrix based on the preprocessed data;

[0010] Step S3: Establish an error data time series characteristic strategy table;

[0011] Step S4: Compare and identify timescale anomalies;

[0012] Step S5: Anomaly location and alarm.

[0013] Preferably, in step S1, the data acquisition and preprocessing includes the following steps:

[0014] Real-time data of the power system is acquired using a distributed phasor measurement unit (PMU).

[0015] A three-level threshold filtering mechanism is used to filter the collected data. The first-level threshold directly removes obviously erroneous data that is out of range; the second-level threshold marks data that is out of range as suspicious values; and the third-level threshold marks data that exceeds the standard deviation range for three consecutive frames.

[0016] Multimodal data fusion technology is introduced to access meteorological data, geographic data, and equipment status data through standardized interfaces;

[0017] An enhanced Kalman filter algorithm is used to fuse multi-source data and establish state equations and observation equations.

[0018] Preferably, in step S2, constructing the temporal relative change matrix includes the following steps:

[0019] For the preprocessed power system data, the relative changes in phasor information are calculated according to the time series, and the magnitudes of voltage, current, and other quantities are expressed using formulas. Calculation, where For time intervals, ;

[0020] For phase angle quantity Calculate the change in angle;

[0021] Fractal analysis techniques are introduced, and an improved box-dimensional method is used to extract fractal features.

[0022] Construct a three-dimensional time series matrix based on PMU number and time window. The matrix elements include relative change values ​​and fractal dimension, and the data is organized using a row-major storage format.

[0023] Preferably, in step S3, establishing the error data time-series characteristic strategy table includes the following steps:

[0024] Construct a training sample library containing 20 typical time-stamped anomalies;

[0025] An unsupervised learning framework is constructed based on Deep Belief Network (DBN), with the network structure being input layer - hidden layer 1 - hidden layer 2 - hidden layer 3 - output layer; the training process adopts a layer-by-layer pre-training and fine-tuning mechanism.

[0026] Strategy Table It is stored in the form of a matrix array, with each exception mode corresponding to a matrix dimension and a real-time matrix. It is consistent and includes three types of statistical features: mean, standard deviation, and extreme value distribution. It also supports the dynamic addition of new anomaly patterns through an incremental learning interface.

[0027] Preferably, in step S4, the comparison and discrimination of timescale anomalies adopts a two-level matching mechanism, including the following steps:

[0028] First, global similarity is calculated, followed by local feature verification. The global similarity calculation uses a dynamic time warping (DTW) and cosine similarity weighted fusion algorithm. For local feature verification, the top 3 candidate anomaly patterns in terms of similarity are selected, and three key indicators—the location of feature mutation points, duration, and rate of change of amplitude—are extracted for secondary verification.

[0029] A dual threshold discrimination mechanism is set.

[0030] A wide-area measurement system data time-stamp anomaly identification system includes:

[0031] The data acquisition and preprocessing module uses the IEC61850-9-2 protocol to communicate with the PMU and configure a multimodal data interface. It also integrates sliding window filtering and Kalman fusion algorithms to clean and process the acquired data from multiple sources, ultimately outputting a high-quality dataset.

[0032] The temporal relative change matrix construction module calculates the relative change of phasor information based on preprocessed data, integrates a fractal feature extraction unit to calculate the box dimension, and then uses an efficient matrix operation library to construct the temporal feature matrix.

[0033] The strategy table creation module is equipped with the DBN deep learning framework. It generates anomaly pattern strategy tables by training on historical data. It also supports incremental learning and updates, which can continuously improve the anomaly pattern library. It integrates visualization tools to realize the graphical display of anomaly patterns for easy and intuitive analysis.

[0034] The comparison and discrimination module deploys the DTW-cosine composite similarity algorithm, which makes full use of the parallel computing power of multi-core processors to improve the efficiency of similarity calculation. It also integrates a decision tree secondary discrimination unit to perform feature re-judgment for fuzzy samples with similarity close to the threshold, so as to ensure the accuracy of time stamp anomaly identification.

[0035] The anomaly location and alarm module is used to link the PMU equipment ledger with the GIS map to achieve spatial location of the anomaly occurrence. It generates alarm information in the IEC61970 standard format to ensure the standardization and compatibility of the information. It also has alarm statistical analysis function, which can explore the patterns of anomalies and make trend predictions, providing strong support for the operation and maintenance decision-making of the power system.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] This invention constructs a temporal relative change matrix that integrates fractal features and compares it with a predefined strategy table using a composite similarity calculation method combining DTW and cosine similarity. This enables rapid and accurate determination of whether data timescales are abnormal and the type of abnormality. Compared with traditional methods, this significantly improves the speed and accuracy of judgment and reduces the false alarm rate of WAMS-based intelligent fault diagnosis and accident handling auxiliary decision-making systems.

[0038] This invention can promptly and accurately identify data time-stamp anomalies, helping to ensure the reliability of various power system applications based on WAMS data. It enables dispatchers to accurately assess the real-time status of the power grid and make correct decisions, thereby improving the safe and stable operation of the power system. By introducing multimodal data fusion, GIS technology, and intelligent analysis algorithms, it can more comprehensively monitor the operating status of the power system, promptly identify potential safety hazards, and provide stronger guarantees for the safe operation of the power system.

[0039] The system designed in this invention features clearly defined functions for each module, exhibiting excellent flexibility and scalability. The strategy table creation module can continuously update and optimize the strategy table based on newly emerging time-stamped anomalies. The data acquisition and preprocessing module can easily connect to more PMU devices and other types of monitoring data to adapt to the needs of wide-area measurement systems of varying scales and complexities. Furthermore, the system modules employ standardized interfaces and communication protocols, facilitating system integration and upgrades. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] like Figure 1As shown, a method for identifying time-stamp anomalies in data from a wide-area measurement system includes the following steps:

[0043] Step S1, Data Acquisition and Preprocessing:

[0044] Step S11: Real-time data of the power system is collected using a distributed phasor measurement unit (PMU). The PMU is deployed at 220kV and above substations, key transmission line nodes and new energy connection gateways. The collection period is no more than 40ms and the synchronization accuracy reaches ±1μs.

[0045] Step S12: The real-time data collected covers 13 types of basic phasor information, including three-phase voltage amplitude and phase angle, three-phase current amplitude and phase angle, active power, reactive power, and system frequency, as well as UTC time-stamped data accurate to the millisecond level.

[0046] Step S13: Adopt a three-level threshold filtering mechanism: The first-level threshold is set to ±20% of the rated value (e.g., the voltage threshold for a 110kV system is 88-132kV), and obviously erroneous data that exceeds the range is directly removed; the second-level threshold is ±10% of the rated value, and data that exceeds the range is marked as suspicious values; the third-level threshold is the 3σ criterion, and data that exceeds the standard deviation range for 3 consecutive frames is marked.

[0047] Step S14: Innovatively introduce multimodal data fusion technology, and access meteorological data (including ambient temperature, humidity, wind speed, precipitation intensity, sampling period of 1 minute), geographic data (including static parameters such as altitude, terrain slope, and route direction, updated quarterly) and equipment status data (including PMU operating temperature and communication link bit error rate) through standardized interfaces.

[0048] Step S15: Use the enhanced Kalman filter algorithm to fuse multi-source data, establish state equations and observation equations, where the state vector dimension is 16 (including 13 types of electrical quantities + 3 types of environmental parameters), and the process noise covariance matrix... The initial value is set as a diagonal matrix. Observation noise covariance matrix By dynamically adjusting the real-time data fluctuation characteristics, the output frequency of the fused data remains at 25 frames / s.

[0049] Among them, the state equation It is the state-space model expression of a discrete-time linear dynamic system, used in the identification of time-scaled anomalies in wide-area measurement system data to describe the change law of system state over time.

[0050] in, Represents the system at discrete time points The state vector contains key state information of the system at that moment, such as measured data like voltage amplitude and phase angle.

[0051] That is, the system at the previous moment The state vector;

[0052] The state transition matrix determines the transition relationship of the system from the previous state to the current state, reflecting the dynamic characteristics of the system itself.

[0053] Is the system at any time The input vector represents the external influence on the system, such as control commands.

[0054] It is the input matrix, used to describe the input vector. Regarding system status The mode of action;

[0055] Indicates the system at time 10:00 The process noise vector reflects the impact of uncertainties such as modeling errors and external disturbances on the system state.

[0056] Observation equations It is a linear observation equation commonly used in signal processing, system modeling, and other fields.

[0057] in,

[0058] Indicates at time The observation vector is the actual measurement data that can be obtained;

[0059] Indicates at time The system state vector represents the true state of the system and is usually not directly measurable.

[0060] It is the observation matrix, which describes the system state vector. to observation vector The mapping relationship is used to transform the system state into an observable form through matrix multiplication;

[0061] Indicates at time The observation noise vector reflects the errors and interferences present in the measurement process, such as the influence of sensor measurement errors and environmental noise on the measurement results.

[0062] Step S2: Construct the time-series relative change matrix:

[0063] Step S21: For the preprocessed power system data, calculate the relative changes in phasor information according to the time series, and apply the formulas to the amplitude quantities such as voltage and current. Calculation, where The time interval (initially set to 10 frames, or 0.4 seconds). ;

[0064] Indicates the first From the point in time to the The relative rate of change of voltage at each time point;

[0065] For the first Voltage values ​​at each time point;

[0066] For the first Each time point has elapsed time intervals The formula calculates the voltage value after the initial voltage by dividing the voltage difference between two adjacent time points by the initial voltage. This yields the relative change in voltage over the time interval, which helps in analyzing the dynamic trend of voltage changes.

[0067] For phase angle quantity Calculate the change in angle;

[0068] in, This formula is used to calculate the first... Each phase value after one cycle Phase change after The specific meaning is:

[0069] Indicates the first Each phase value passes through a period The subsequent phase value, For the initial first Each phase value.

[0070] Calculate the absolute value of the difference between the phase values ​​at two different times to obtain the magnitude of the phase change.

[0071] This means taking the calculated phase change magnitude modulo 360° to ensure the final phase change amount is accurate. Within the range of 0° to 360°.

[0072] Step S22: Introduce fractal analysis technology and extract fractal features using the improved box dimension method: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The relative change series is divided into The side length is The box, The value ranges from 0.01 to 0.1 (step size 0.01), and the number of boxes covering non-zero data is calculated at each scale. ;

[0073] Fitting the double logarithmic curve using the least squares method ,in,

[0074] Indicated by scale Number of boxes required to cover the target set The natural logarithm;

[0075] Fractal dimension describes the irregularity or complexity of a set. The larger the value, the higher the complexity of the object;

[0076] It is a measurement scale The natural logarithm of the reciprocal reflects the relationship between measurement accuracy and coverage quantity;

[0077] The constant term represents the intercept of the formula and does not affect the fractal dimension. The calculation.

[0078] Step S23: Construct a three-dimensional time series matrix according to the PMU number (1-N) and the time window (dividing each hour into 9000 time intervals). This expression represents It is a three-dimensional matrix, where Represents the real number field. The first dimension of the matrix is ​​9000, the second dimension is 2, and the third dimension is 2. The matrix elements include the relative change value (6 decimal places) and the fractal dimension (3 decimal places). The data is organized in row-major storage format.

[0079] Step S3: Establish an error data time series characteristic strategy table:

[0080] Step S31: Construct a training sample library containing 20 typical time-stamped anomalies, covering scenarios such as: GPS signal loss (four durations: 1s / 3s / 5s / 10s), clock jumps (three amplitudes: ±1s / ±500ms / ±100ms), communication delays (three levels of delay: 50ms / 100ms / 200ms), data duplication (2-5 consecutive frames of duplication), time-stamped reversal (negative time difference between adjacent frames), and sampling frequency offset (±20% deviation).

[0081] Step S32: Construct an unsupervised learning framework based on Deep Belief Network (DBN). The network structure is: input layer (dimension equal to the number of features) - hidden layer 1 (512 nodes) - hidden layer 2 (256 nodes) - hidden layer 3 (128 nodes) - output layer (20 anomaly labels).

[0082] Step S33: The training process adopts a layer-by-layer pre-training and fine-tuning mechanism: In the pre-training stage, the contrastive divergence algorithm is used with a learning rate of 0.01, a momentum factor of 0.9, and each RBM is trained for 50 epochs; in the fine-tuning stage, the Adam optimizer is used with a batch size of 64, a total number of iterations of 500, and a convergence threshold of 1e-5 for the cross-entropy loss function.

[0083] Step S34, Strategy Table It is stored in the form of a matrix array, with each exception mode corresponding to a matrix dimension and a real-time matrix. Consistent, including three types of statistical features: mean, standard deviation, and extreme value distribution, and supports the dynamic addition of new anomaly patterns through the incremental learning interface (the sample size of the new pattern is no less than 500 groups).

[0084] Step S4: Compare and identify timescale anomalies:

[0085] Step S41, Real-time Matrix With strategy table The comparison adopts a two-level matching mechanism: first, global similarity is calculated, and then local feature verification is performed;

[0086] Step S42: Global similarity calculation uses a dynamic time warping (DTW) and cosine similarity weighted fusion algorithm: the DTW algorithm sets the window constraint to 50 (i.e., the maximum time offset is 50 frames) and calculates the path cost. ;

[0087] Cosine similarity is calculated by taking the cosine of the angle between vectors.

[0088] ,

[0089] The formula It represents a vector sum vector The cosine similarity. Among them, the molecules... For vectors with vector The dot product (also called the inner product) reflects the degree of coordination between two vectors in direction; the denominator It is a vector The modulus (length) and vector The product of the moduli (lengths) is used to normalize the molecules. The range of values ​​for cosine similarity is... The closer the value is to 1, the more similar the directions of the two vectors are; the closer the value is to -1, the more opposite the directions are; and a value of 0 indicates that the two vectors are orthogonal (perpendicular). In the identification of timescale anomalies in wide-area measurement systems, cosine similarity can be used to measure the degree of similarity between different data feature vectors, thereby helping to determine whether there are anomalies in the timescale.

[0090] Overall similarity ,

[0091] in, It is a comprehensive index, calculated based on cosine similarity. And metrics calculated based on Dynamic Time Warping (DTW) The weighted summation yields results with weights of 0.6 and 0.4, respectively. This indicates that in the identification of time-scaled anomalies in wide-area measurement system data, both cosine similarity and dynamic time warping methods are considered simultaneously, and they are fused together with specific weights to measure data characteristics or determine anomalies.

[0092] Indicators of comprehensive indicators It is based on a metric of cosine similarity. Compared with metrics based on dynamic time warping It is a weighted combination with weights of 0.6 and 0.4. In the scenario of identifying data timescale anomalies in wide-area measurement systems, this weighting method comprehensively considers the results obtained from two different calculation methods to more accurately determine whether there are anomalies in the data timescale.

[0093] Step S43: Local Feature Verification. For the top 3 candidate abnormal patterns in terms of similarity, three key indicators are extracted for secondary verification: the location of feature mutation points, duration, and amplitude change rate.

[0094] Step S44: Set up a dual threshold discrimination mechanism: threshold (Triggered an anomaly warning) threshold (Confirm the exception type), when The exception type is determined directly when... The manual review process will be initiated at that time.

[0095] Step S5, Anomaly Location and Alarm:

[0096] Step S51: Based on the PMU equipment ledger (including unique equipment ID, installation location latitude and longitude, and name of the substation / line to which it belongs) and data time stamp, construct a spatiotemporal index table to achieve abnormal location positioning accuracy at the line tower level and time positioning accuracy at the 10ms level.

[0097] Step S52: GIS visualization adopts a layered display technology: the bottom layer is a geographic base map (scale 1:50000), the middle layer is a power grid topology map (including substation, line, and tower identification), and the top layer is anomaly markers (different colors are used to distinguish the anomaly level: yellow warning / red alarm).

[0098] Step S53: The alarm information is encapsulated in the IEC61970 standard format, including: anomaly code (10-digit code, the first 4 digits identify the anomaly type), occurrence time (UTC time accurate to milliseconds), PMUID, location coordinates (WGS84 coordinate system), characteristic parameters (anomaly duration, maximum deviation value), and confidence level (0-100%), totaling 6 core contents.

[0099] Step S54: Alarm push adopts a dual-channel mechanism: it is pushed to the main station system in real time through the power dispatch data network (delay ≤1s), and at the same time pushed to the operation and maintenance personnel terminal through the SMS gateway (delay ≤30s), supporting the confirmation, forwarding and archiving of alarm information.

[0100] like Figure 1 As shown, a wide-area measurement system data time-stamp anomaly identification system includes:

[0101] Data acquisition and preprocessing module:

[0102] At the communication interface level, we strictly follow the IEC61850-9-2 protocol to build a stable and reliable communication link with the PMU, and continuously and efficiently collect real-time data of the power system at a frequency of 25 frames / s to ensure the timeliness and completeness of data acquisition.

[0103] To address the need for multimodal data access, we have carefully configured standardized and highly versatile multimodal data interfaces to seamlessly connect with diverse environmental data sources such as meteorology and geography, breaking down data barriers and achieving data fusion and convergence.

[0104] Internally, it integrates sliding window filtering and Kalman fusion algorithms to form a data processing pipeline. The sliding window filtering flexibly adjusts the window width based on the data fluctuation characteristics to efficiently filter out high-frequency noise interference; the Kalman fusion algorithm, based on state equations and observation equations, performs deep fusion of multi-source data, optimizes data quality, and ultimately outputs a high-quality dataset that meets the needs of subsequent analysis.

[0105] Time-series relative change matrix construction module:

[0106] Based on the high-quality data output from the data acquisition and preprocessing module, we deeply explore the dynamic characteristics of the data. Through efficient algorithms, we calculate the relative changes in phasor information in parallel, and precisely calculate the amplitude and phase angle using dedicated formulas, capturing subtle changes in the data over time.

[0107] An advanced fractal feature extraction unit is introduced, and a detailed analysis of the relative change sequence is performed based on the improved box dimension method. By reasonably setting the range and step size of the box side length (epsilon), the fractal dimension features of the data are accurately extracted, providing a unique perspective for subsequent analysis.

[0108] Using an optimized, high-efficiency matrix operation library, a three-dimensional time-series feature matrix is ​​constructed based on the PMU number and time window. This module features intelligent adaptive time interval adjustment, which can dynamically adjust the time interval within the range of 0.1-1 seconds according to the frequency of changes in the power system's operating status. It also supports parallel processing of multiple types of phasor data, significantly improving computational efficiency and data processing capabilities.

[0109] Strategy table creation module:

[0110] Equipped with the advanced DBN deep learning framework, which includes 3 layers of Restricted Boltzmann Machine (RBM) and 1 layer of classification, it constructs a powerful abnormal pattern learning and recognition system.

[0111] By utilizing historical data for deep training and employing a layer-by-layer pre-training and fine-tuning mechanism, the network parameters are continuously optimized to generate a detailed and accurate anomaly pattern policy table. Simultaneously, it supports incremental learning and updating, automatically updating the policy table when new anomaly data appears, continuously improving the anomaly pattern library.

[0112] It integrates a wealth of visualization tools, which can display anomaly patterns in a variety of intuitive graphical ways such as line charts, bar charts, and heatmaps, making it easy for users to quickly understand the characteristic distribution of different anomaly patterns and providing strong support for data analysis and decision-making.

[0113] Comparison and discrimination module:

[0114] The independently developed DTW-cosine composite similarity algorithm is deployed to fully explore the parallel computing potential of multi-core processors. Through reasonable task allocation and scheduling, the efficiency of similarity calculation is greatly improved, and the rapid comparison between real-time matrices and policy tables is achieved.

[0115] The integrated decision tree secondary discrimination unit performs in-depth re-judgment on fuzzy samples with similarity close to the threshold from multiple dimensions such as the location of feature mutation points, duration, and amplitude change rate, effectively avoiding misjudgment and ensuring high accuracy in time stamp anomaly identification.

[0116] Anomaly location and alarm module:

[0117] By deeply linking key information such as device ID and installation location in the PMU equipment ledger with the GIS map, an efficient spatiotemporal index table is constructed. Based on this, the location of anomalies can be quickly and accurately spatially located, achieving high-precision positioning at the line tower level; combined with data time stamps, time positioning accuracy reaches 10ms level, precisely pinpointing the time of anomaly occurrence.

[0118] Alarm information is generated strictly in accordance with the IEC 61970 standard format to ensure standardization and compatibility across different systems. Alarm information includes core elements such as anomaly code, occurrence time, PMUID, location coordinates, characteristic parameters, and confidence level, providing maintenance personnel with comprehensive and accurate anomaly information.

[0119] It features powerful alarm statistical analysis capabilities, enabling multi-dimensional statistical analysis of abnormal data, including the frequency, duration, and scope of impact of anomalies. Through in-depth mining of historical anomaly data, it predicts anomaly trends, providing a scientific basis and data support for power system operation and maintenance decisions.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying time stamp anomaly of wide area measurement system data, characterized in that, The method comprises the following steps: Step S1, data acquisition and preprocessing; Step S2, constructing a time series relative change matrix according to the preprocessed data; Step S3, establishing an error data time series feature strategy table; Step S4, comparing and discriminating time scale anomalies; Step S5, anomaly positioning and alarm. 2.The method and system for identifying time tag anomaly of wide-area measurement system data according to claim 1, wherein, In step S1, the data acquisition and preprocessing comprises the following steps: Real-time data of the power system is collected by using a distributed phasor measurement unit (PMU); A three-level threshold filtering mechanism is adopted to filter the collected data; wherein, the first-level threshold directly eliminates obviously erroneous data that exceeds the range; the second-level threshold marks the data that exceeds the range as suspicious values; the third-level threshold marks the data that exceeds the standard deviation range for three consecutive frames; A multi-modal data fusion technology is introduced, and meteorological data, geographical data and equipment state data are accessed through a standardized interface; An enhanced Kalman filtering algorithm is used for multi-source data fusion, and a state equation and an observation equation are established. 3.The method and system for identifying abnormal time stamp of wide area measurement system data according to claim 1, wherein, In step S2, the construction of the time series relative change matrix comprises the following steps: For the pretreated power system data, the relative change amount of the phasor information is calculated according to the time sequence, and the formula is used for the amplitude value of the voltage, the current and the like is calculated, wherein is a time interval The phase angle quantity is calculated as The angle change is calculated A fractal analysis technology is introduced, and an improved box dimension method is used to extract fractal features; Constructing three-dimensional time sequence matrix according to PMU number and time window The matrix elements include relative change value and fractal dimension, and the data is organized in row-first storage format. 4.The method and system for identifying abnormal time stamp of wide area measurement system data according to claim 1, wherein, In step S3, the establishment of the error data time series feature strategy table comprises the following steps: A training sample library containing 20 typical time scale anomalies is constructed; An unsupervised learning framework is constructed based on a deep belief network (DBN), and the network structure is input layer-hidden layer 1-hidden layer 2-hidden layer 3-output layer; wherein, the training process adopts a layer-by-layer pre-training and fine-tuning mechanism; Policy table Stored in the form of matrix array, each anomaly pattern corresponds to the dimension of the matrix and the real-time matrix Consistent, containing three types of statistical features including mean, standard deviation and extreme value distribution, supporting dynamic addition of new anomaly patterns through incremental learning interface.

5. The method and system for identifying time stamp anomalies in wide area measurement system data according to claim 1, wherein, In step S4, the comparison and discrimination of time scale anomalies adopt a two-level matching mechanism, which comprises the following steps: First, global similarity calculation is performed, and then local feature verification is performed; wherein, the global similarity calculation adopts a dynamic time warping (DTW) and cosine similarity weighted fusion algorithm; the local feature verification extracts three key indicators, i.e., feature mutation point position, duration and amplitude change rate, for secondary verification for the top 3 candidate anomaly patterns in terms of similarity; A double threshold discrimination mechanism is set.

6. A wide area measurement system data timestamp anomaly identification system, characterized in that, It comprises: A data acquisition and preprocessing module, which realizes communication docking with the PMU and configures a multi-modal data interface by adopting the IEC61850-9-2 protocol, and integrates a sliding window filtering and Kalman fusion algorithm to sequentially clean and multi-source fuse the collected data, and finally outputs a high-quality data set; A time series relative change matrix construction module, which calculates the relative change of phasor information based on the preprocessed data, integrates a fractal feature extraction unit to calculate the box dimension, and then uses an efficient matrix operation library to construct a time series feature matrix; A strategy table establishment module, which carries a DBN deep learning framework, generates an anomaly pattern strategy table through training of historical data, supports incremental learning update, can continuously improve the anomaly pattern library, and integrates a visualization tool for graphical display of the anomaly pattern for intuitive analysis; The comparative discrimination module deploys a DTW-cosine composite similarity algorithm, fully utilizes the parallel computing capability of the multi-core processor, improves the similarity calculation efficiency, and integrates a decision tree secondary discrimination unit, so that the feature of a fuzzy sample with a similarity close to a threshold value is judged again, to ensure the accuracy of the time scale anomaly identification. The abnormal positioning and alarm module is used for associating PMU equipment accounts with GIS maps, realizing spatial positioning of the abnormal position, generating alarm information in the IEC 61970 standard format, ensuring the standardization and compatibility of the information, and configuring an alarm statistical analysis function, which can mine the rules of abnormal occurrence and make trend prediction, to provide strong support for the operation and maintenance decision of the power system.

Images